Abstract
Adjoint optimization is an exciting and fast growing field that has many applications in the automotive industry. We focus on the usage of adjoint for the optimization of aerodynamic performance. While adjoint methods have come to the attention of mainstream CFD through inclusion in prominent commercial codes, most of the available tools are severely limited, precluding productive use in this field. We detail a methodology that is based on the continuous adjoint method and is implemented in an Open Source framework. While more mathematically demanding in terms of its derivation, the continuous adjoint method requires orders of magnitude fewer resources without sacrificing significant accuracy. In this paper continuous adjoint methods are used for calculating gradients of aerodynamic objective functions (drag, lift, moments etc.) in applications with a huge number of design variables. Methodologies that accept either steady-state RANS(1)(2)(3) or time averaged DES(4)(6) as primal flow input are outlined and extensions to improve the accuracy of previously published methods are detailed. Finally, a novel methodology, based on volumetric B-splines to translate the surface sensitivities produced by the adjoint into optimized shapes, is introduced and showcased.
